Panel data instrumental variable analysis

[Mattt Wordingham]

Hello,

I am trying to conduct a panel data IV analysis using the xtivreg command in Stata 15.1.

I am trying to instrument an institutional quality variable, but my instruments are showing 0 coefficients and omitted standard error in the first stage regression. My institutions variable is then suffering from the same problem in the second stage regression.

Is there an error with my code or another problem I am missing?

Here is a copy of my state output:

xtivreg rgdppc nr educ gcf fdi trade (inst = slaveex abslat frac) i.year, fe vce(rob
> ust) first

First-stage within regression

Fixed-effects (within) regression Number of obs = 435
Group variable: ID Number of groups = 36

R-sq: Obs per group:
within = 0.1663 min = 1
between = 0.1509 avg = 12.1
overall = 0.1774 max = 18

F(21,35) = .
corr(u_i, Xb) = 0.2078 Prob > F = .

(Std. Err. adjusted for 36 clusters in ID)
------------------------------------------------------------------------------
| Robust
inst | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
nr | -.0054985 .0022037 -2.50 0.017 -.0099723 -.0010246
educ | 3.53e-08 8.07e-08 0.44 0.665 -1.29e-07 1.99e-07
gcf | 1.46e-12 3.03e-12 0.48 0.632 -4.69e-12 7.61e-12
fdi | -.0014125 .0004381 -3.22 0.003 -.0023019 -.000523
trade | .0006766 .0003794 1.78 0.083 -.0000937 .0014469
|
year |
1998 | -.0521823 .0469836 -1.11 0.274 -.1475641 .0431995
2000 | -.0669626 .0501403 -1.34 0.190 -.1687528 .0348275
2002 | -.0527042 .053657 -0.98 0.333 -.1616338 .0562254
2003 | -.0308541 .0483765 -0.64 0.528 -.1290636 .0673555
2004 | -.0741527 .0489266 -1.52 0.139 -.1734789 .0251735
2005 | -.0653169 .0529847 -1.23 0.226 -.1728815 .0422477
2006 | -.0325287 .0495982 -0.66 0.516 -.1332184 .068161
2007 | -.0233353 .0543166 -0.43 0.670 -.1336037 .0869332
2008 | -.0397947 .0632159 -0.63 0.533 -.1681298 .0885404
2009 | -.0483473 .060147 -0.80 0.427 -.1704523 .0737577
2010 | -.0707773 .0741316 -0.95 0.346 -.2212725 .0797179
2011 | -.0596762 .0716028 -0.83 0.410 -.2050377 .0856853
2012 | -.0622497 .0695302 -0.90 0.377 -.2034036 .0789042
2013 | -.0864571 .0808348 -1.07 0.292 -.2505606 .0776463
2014 | -.0626817 .0788548 -0.79 0.432 -.2227654 .0974021
2015 | -.0488885 .0902416 -0.54 0.591 -.2320887 .1343118
2016 | -.0550498 .090986 -0.61 0.549 -.2397611 .1296615
|
slaveex | 0 (omitted)
abslat | 0 (omitted)
frac | 0 (omitted)
_cons | -.4706296 .0777993 -6.05 0.000 -.6285707 -.3126886
-------------+----------------------------------------------------------------
sigma_u | .56028398
sigma_e | .14279912
rho | .9390039 (fraction of variance due to u_i)
------------------------------------------------------------------------------

Fixed-effects (within) IV regression Number of obs = 435
Group variable: ID Number of groups = 36

R-sq: Obs per group:
within = 0.3396 min = 1
between = 0.0026 avg = 12.1
overall = 0.0005 max = 18


Wald chi2(22) = 55586.45
corr(u_i, Xb) = -0.3575 Prob > chi2 = 0.0000

(Std. Err. adjusted for 36 clusters in ID)
------------------------------------------------------------------------------
| Robust
rgdppc | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
inst | 0 (omitted)
nr | -3.797156 3.650058 -1.04 0.298 -10.95114 3.356827
educ | -.0005414 .0002069 -2.62 0.009 -.000947 -.0001359
gcf | 3.83e-09 9.95e-09 0.39 0.700 -1.57e-08 2.33e-08
fdi | -2.216131 .7564366 -2.93 0.003 -3.698719 -.7335423
trade | -2.800947 1.993689 -1.40 0.160 -6.708505 1.10661
|
year |
1998 | -161.7154 128.7196 -1.26 0.209 -414.0013 90.5704
2000 | 30.72823 166.213 0.18 0.853 -295.0433 356.4997
2002 | 156.0356 204.7612 0.76 0.446 -245.2891 557.3602
2003 | 202.2323 192.8986 1.05 0.294 -175.842 580.3065
2004 | 287.3992 212.6768 1.35 0.177 -129.4396 704.2381
2005 | 335.4165 223.2142 1.50 0.133 -102.0752 772.9083
2006 | 423.1018 264.1035 1.60 0.109 -94.53155 940.7352
2007 | 531.9081 295.779 1.80 0.072 -47.80805 1111.624
2008 | 589.2614 325.3914 1.81 0.070 -48.49407 1227.017
2009 | 587.6815 321.1732 1.83 0.067 -41.80646 1217.169
2010 | 696.0536 359.4717 1.94 0.053 -8.498037 1400.605
2011 | 741.3152 369.3038 2.01 0.045 17.49293 1465.137
2012 | 918.499 407.7731 2.25 0.024 119.2784 1717.72
2013 | 1006.988 434.1838 2.32 0.020 156.0036 1857.973
2014 | 1058.819 443.5378 2.39 0.017 189.5006 1928.137
2015 | 1087.012 452.6947 2.40 0.016 199.7465 1974.277
2016 | 1116.399 459.1283 2.43 0.015 216.5242 2016.274
|
_cons | 2021.848 247.6014 8.17 0.000 1536.558 2507.138
-------------+----------------------------------------------------------------
sigma_u | 2518.3393
sigma_e | 373.74069
rho | .97844982 (fraction of variance due to u_i)
------------------------------------------------------------------------------
Instrumented: inst
Instruments: nr educ gcf fdi trade 1998.year 2000.year 2002.year 2003.year
2004.year 2005.year 2006.year 2007.year 2008.year 2009.year
2010.year 2011.year 2012.year 2013.year 2014.year 2015.year
2016.year slaveex abslat frac
------------------------------------------------------------------------------

[Carlo Lazzaro]

Matt:
are your instruments time-invariant?

[Mattt Wordingham]

Hi Carlo,

Thanks for your reply.

That's right, my instruments are time-invariant.

Does that mean I have to change some of the code?

[Carlo Lazzaro]

Matt:
the -fe- specification wipes out time-invariant predictors/instruments.
There's nothing you can about that, but change the instruments (if feasible).

[Mattt Wordingham]

Thanks very much for your help Carlo.

[Mattt Wordingham]

Hi Carlo,

Thanks for your help last week.

I've changed my instruments now.

Is there a way to test for enodgeneity and overidentification after running the following command:

xtivreg rgdppc nr educ gcf fdi trade i.year (INST = l5INST), fe vce(robust) first

I've tried xtoverid, but get this error message:

xtoverid
2003b: operator invalid
r(198);

I've also tried a hausman test with this code:

xtreg rgdppc nr INST educ gcf fdi trade i.year, fe robust
estimates store xtreg
xtivreg rgdppc nr educ gcf fdi trade i.year (INST = l5INST), fe vce(robust) first
estimates store xtivreg
hausman xtreg xtivreg, sigmamore

This gives me this error:

hausman cannot be used with vce(robust), vce(cluster cvar), or p-weighted data
r(198);


Any help would be much appreciated

Kind regards,

Matt

[Mattt Wordingham]

Estat endogenous also doesn't work

[Carlo Lazzaro]

Matt:
the user-written command -xtoverid- is a bit too old to support -fvvarlist- notation.
The fix is to re-write your first code as.

xi: xtivreg rgdppc nr educ gcf fdi trade i.year (INST = l5INST), fe vce(robust) first

As far as your second code is concerned, you should use -xtoverid- and -xi:- again: unfortunately, -hausman- does not support non-default standard errors.